NVIDIA Research builds AI tools to convert 2D image series into impressive 3D models


NVIDIA researchers Has created a new reverse rendering pipeline, NVIDIA 3D MoMaWhich allows users to recreate a 3D model of an object or even a series of stills in a scene.

The ability to reconstruct a series of still photos on a 3D model has long been a sacred grail to integrate computer vision and computer graphics, ‘said David Lube, vice president of graphics research at NVIDIA. By formulating each part of the reverse rendering problem as a GPU-accelerated differentiable component, the NVIDIA 3D MoMa rendering pipeline can quickly create 3D objects using modern AI equipment and NVIDIA GPU’s raw computational horsepower that manufacturers can import, edit and edit. Expand on existing tools without limitation, ‘he said.

These five models were created using NVIDIA 3D MoMa. Credit: NVIDIA

The most useful 3D objects are created using a triangle mesh and textured materials. Such objects can easily be dropped into game engines, 3D modulators, and film renderers. However, these models are usually created by hand or using techniques such as photogrammetry. In both cases, there is the promise of a significant time. On the other hand, NVIDIA 3D MoMa uses a single NVIDIA Tensor Core GPU to create a triangular lattice model in one hour. The output is compatible with existing 3D graphics engines and modeling tools.

Reconstruction consists of three primary components: a 3D mesh model, materials and lighting. The mesh is skeleton-like, or according to NVIDIA, ‘a 3D-shaped papier-m মডেলch মডেল model made from triangles.’ The material is a 2D texture that goes into the 3D mesh model. 3D MoMa then creates an approximate lighting model, which manufacturers can change later.

To showcase 3D MoMa and celebrate the rich jazz history in New Orleans, where Paper 3D MoMa will be presented on the back Conference on Computer Vision and Pattern Recognition, NVIDIA captured about 100 images of each of the five jazz band instruments from different angles. The trumpet, trombone, saxophone, drum set and clarinet were then modeled using wax. You can see the results above.

Earlier this year, NVIDIA Research showed instant NERF, which could turn 2D images into 3D scenes ‘in seconds’. Not only that, so far in 2022, we have also seen the groundbreaking achievements of NVIDIA in neural graphics training and an impressive new NVIDIA canvas update. There is no doubt that AI offers some great benefits to creators, especially when it comes to saving time creating artwork and models.

Source link